Agentverkstad
Menu

Article · Last updated 11 November 2026

Why many AI pilots never reach daily use

An AI pilot reaches daily use when someone owns it, when it starts with a single workflow and when the people who will use it have chosen it. The reason pilots stall usually sits with people and decisions, and seldom in the tool itself.

The tool is seldom what gets stuck

When an AI project stalls, the eye often goes to the technology. That is understandable, because the technology is what you can see. But the reasons we most often see are others: it is unclear who owns the question, people worry about being replaced, different people hold different pictures of what AI is for, or nobody has time to try.

That is good news, because those reasons can be mapped and dissolved. Swapping a tool seldom solves any of them.

Start with a single workflow

Many pilots try to prove too much. They are meant to save time in finance, improve customer meetings and change how management works, all at once. Then there is nothing to see in everyday work, and nobody can say whether it works.

One workflow is enough: something that recurs and that today requires someone to move information between systems by hand. When it works it shows, and then there is something to build on.

A stop point after every step

After every step we stop and you see what works. You move on only when you want to. That makes the pilot easier to say yes to, because nobody commits to more than the next step, and it keeps mistakes small and fixable.

The price of the next step is set before it starts. Then there are no surprises after a good result.

Someone has to own it

A pilot with no named owner on your side has nobody who decides when it is done, nobody who takes questions from the staff and nobody who decides on the next step. The owner needs the mandate to say what happens next week, and needs no technical background.

That is also why the question belongs to both HR and the AI team. One defined team and one named owner give the question a home.

Measure three things

We agree three measures before the pilot starts: use, time saved and a short pulse of confidence. Use shows whether the agent is part of everyday work. Time saved shows whether it frees something. Confidence shows whether the people around it dare to rely on it.

Without measures the stop point becomes a matter of opinion. With three measures it becomes a conversation about what the numbers show.

Let the team choose

An agent that the team has chosen to use lives longer than one introduced from above. That is why the first design question is how people meet the agent: short messages in the right language, in tools they already have, with no new login.

When an AI project has stalled it can help to first map who decides, what feels reasonable to change and where it has locked up. That is the work we do in AI Adoption, together with 21 Dimensions.

How to start

Appoint an owner, choose a workflow that recurs, agree three measures and a team that gets to try. Then book a first conversation. It is 30 minutes and free of charge, and you tell us where the time disappears.

That is enough to start. A pilot does not need to be big to show whether it reaches daily use, and it can be stopped after the first step if it does not.

Read on

Tell us where the time goes.

Book a first call straight away, 30 minutes and free of charge. Or leave a few lines and we will get back to you.

Fetching available times …

Other ways in